首页|期刊导航|环球中医药|基于动态不确定因果图的名老中医辨治模型构建与应用——以朱仁康湿疹医案为例

基于动态不确定因果图的名老中医辨治模型构建与应用——以朱仁康湿疹医案为例OA

Interpretable modeling and application of the renowned traditional Chinese medicine practitioner's syndrome differentiation and treatment model based on dynamic uncertain causality graph——illustrated by Zhu Renkang's eczema medical cases

中文摘要英文摘要

目的 基于动态不确定因果图(dynamic uncertain causality graph,DUCG)理论,构建中医皮外科专家朱仁康湿疹诊疗经验的智能诊断模型,探索 DUCG 在名老中医经验传承中,对于提高证型诊断精度、实现辨证思维可视化及增强可解释性推理的支持作用.方法 系统梳理朱仁康相关书籍、文献、医案,提炼并获取湿疹常见18 类中医证型、71 项症状体征及8 类风险因素,并将上述关键信息与数据映射为逻辑变量,构建湿疹 DUCG 诊断模型.整理朱仁康亲诊湿疹医案库,随机抽取30 例医案用于模型参数校正与优化,以提高诊断精度.使用优化后的 DUCG 模型对 120 例独立测试集医案进行3 轮证型验证,并从诊断准确性、输出稳定性及诊断可解释性3 个维度,与通用大语言模型ChatGPT 5.1、DeepSeek-R1,中医垂类大语言模型华佗GPT Ⅱ,及3 名本学派不同水平临床医生进行对比分析.结果 DUCG 模型的诊断准确率为78.33%,高于医生平均水平(72.50%)、ChatGPT 5.1(37.50%)、DeepSeek-R1(33.06%)和华佗 GPT Ⅱ(28.61%);在稳定性方面,DUCG 模型3 轮结果一致,标准差为0,优于大语言模型及医生组.同时,DUCG 各证型子图与病机简化图可清晰展示诊断依据与推理路径,具备诊断可解释性.结论 本研究构建的朱仁康湿疹 DUCG 诊断模型在准确性、稳定性及可解释性上均展现出优势,为名老中医经验的可视化阐释、结构化保存与智能化传承提供了一种有效方案.

Objective Based on the theory of the Dynamic Uncertain Causality Graph(DUCG),this study aims to construct an intelligent diagnostic model encapsulating the clinical experience of the renowned traditional Chinese medicine(TCM)dermatologist,Dr.Zhu Renkang,in the diagnosis and treatment of eczema.This paper explores the utility of DUCG in the preservation of distinguished veteran TCM practitioners'expertise,specifically focusing on improving the accuracy of syndrome differentiation,visualizing dialectical thinking,and enhancing interpretable reasoning.Methods Relevant books,literature,and medical records of Zhu Renkang were systematically reviewed to extract and obtain 18 common TCM syndrome types,71 symptoms and signs,and 8 categories of risk factors related to eczema;these key information and data were mapped into DUCG logical variables to construct a DUCG model for eczema pattern differentiation.A database of eczema medical cases personally treated by Zhu Renkang was compiled,from which 30 cases were randomly selected for model parameter calibration and optimization to improve diagnostic precision.The optimized DUCG model was used to conduct three rounds of syndrome verification on 120 independent test cases,and a comparative analysis was conducted with the general large language model(LLM)ChatGPT 5.1,DeepSeek-R1,the TCM specialized LLM Huatuo GPT Ⅱ,and three clinicians of the same school with different levels of clinical expertise.Results The diagnostic accuracy of the DUCG model was 78.33%,higher than the average level of doctors(72.50%),ChatGPT 5.1(37.50%),DeepSeek-R1(33.06%),and Huatuo GPT Ⅱ(28.61%).In terms of stability,the results of the DUCG model in three rounds were consistent,with a standard deviation of 0,which was superior to LLMs and the doctor team.Meanwhile,the subgraphs of each syndrome type and the simplified pathogenesis diagrams of DUCG can clearly display the diagnostic basis and reasoning path,achieving complete interpretability.Conclusion The DUCG diagnostic model for Zhu Renkang's eczema constructed in this study has demonstrated advantages in terms of accuracy,stability and interpretability,providing an effective solution for the visual interpretation,structured preservation and intelligent inheritance of the experience of renowned TCM practitioners.

刘明玥;冷学明;骆长永;刘欣茜;张万锞;张湛;宋坪

100091 北京,中国中医科学院西苑医院皮肤科||北京中医药大学研究生院中国科学院大学电子电气与通信工程学院北京中医药大学东方医院感染科100091 北京,中国中医科学院西苑医院皮肤科||北京中医药大学研究生院100091 北京,中国中医科学院西苑医院皮肤科||北京中医药大学研究生院清华海峡研究院大数据中心 AI For Medicine 研究室100091 北京,中国中医科学院西苑医院皮肤科

医药卫生

名医经验人工智能动态不确定因果图朱仁康湿疹中医诊断

experience of famous doctorsartificial intelligencedynamic uncertain causality graphZhu Renkangeczematraditional Chinese medicine diagnosis

《环球中医药》 2026 (3)

420-427,8

北京市高层次创新创业人才支持计划"登峰"项目(G202514020)2025年度北京中医药大学研究生自主科研课题(ZJKT2025001)中央高水平中医医院临床科研基金项目(K2023C14)

10.3969/j.issn.1674-1749.2026.03.003

评论